3D Printed Robots Use AI Machine Learning in U.S. Army Research | The Brink

Applications of AI


See how autonomous robots can create shock-absorbing shapes that humans just can't achieve, and what that means for designing safer helmets, packaging, car bumpers, and more.

engineering

Watch an autonomous robot make a splashA shape with absorbent properties that humans cannot matchWhat does this mean for designing safer helmets, packaging, car bumpers, and more?

In a Boston University engineering lab, a robotic arm drops small plastic objects into a box that sits flush against the floor to catch the drop. One by one, these tiny structures — feather-light cylinders less than an inch tall — are packed into the box. Some are red, some blue, purple, green, and black.

Each object is the result of an experiment in robotic autonomy, where the robot is learning to explore and create the most efficient energy-absorbing shape it has ever seen.

Watch as the robot completes a full experiment, from printing 3D shapes to crushing them under a metal plate and dropping discarded objects into a box.

The robot creates a small plastic structure on a 3D printer, records its shape and size, moves it onto a flat metal surface, then crushes it with the equivalent pressure of an adult Arabian horse standing on a quarter. The robot then measures how much energy the structure absorbed and how its shape changed after compression, recording all the details in a vast database. It then drops the crushed object into a box, wipes the metal plate clean, and prepares to print and test the next piece. It is only slightly different from its predecessor, and its design and dimensions are fine-tuned by the robot's computer algorithms based on all previous experiments. This is the basis of something called Bayesian optimization. With each experiment, the 3D structure becomes better and better at absorbing the impact of being crushed.

The various shapes appear to be melting under the power of MAMA BEAR.

These experiments were made possible thanks to the work of Keith Brown, an associate professor of mechanical engineering at ENG, and his team at the KABlab. Named MAMA BEAR (short for its longer official name, Mechanics of Additively Manufactured Architectures Bayesian Experimental Autonomous Researcher), the robot has evolved since it was first conceptualized by Brown and his lab in 2018. By 2021, the lab had tasked the machine with a quest to create a shape that absorbs energy, a property known as mechanical energy absorption efficiency. This current iteration has been running continuously for more than three years, with more than 25,000 3D printed structures packed into dozens of boxes.

Why so many shapes? The ability to absorb energy efficiently has myriad applications, from cushioning sensitive electronic devices shipped around the world to knee pads and wrist guards for athletes. “From this data library, we can build better car bumpers or better packaging equipment, for example,” Brown says.

To function ideally, a structure must be perfectly balanced: It can't be so strong that it damages what it's supposed to protect, but it needs to be strong enough to absorb impacts. The best structures observed before Mama Bear had an energy absorption efficiency of about 71 percent, Brown says. But on a chilly afternoon in January 2023, Brown's lab watched as the robot achieved 75 percent efficiency, shattering the known record. The results: Nature Communications.

“When we started the project, we weren't sure we'd get this record-breaking shape,” says Kelsey Snapp (ENG'25), a doctoral student in the Brown lab who oversees MAMA BEAR. “Slowly but surely, we kept making progress, bit by bit, and we made a breakthrough.”

evolution

Click the play icon to move each shape and take a closer look at the MAMA BEAR research. Scroll to the end using the yellow arrows to see the record-breaking shapes and watch the evolution of the structure. (Note: these are digital renderings of the real thing. For a more precise technical description, see Nature Communications paper.)

#00069

This initial design, created in May 2021, had an efficiency rating of 44 percent and was just 1.9 centimeters tall. At this point, it's clear that MAMA BEAR had a lot to learn.

#05226

After making thousands of shapes, MAMA BEAR made a small but significant leap: after five months of making shapes at about 44 percent efficiency, it now jumped to 53 percent efficiency. It still has a long way to go to break the record.

#07214

The experiment produced a relatively small increase in efficiency (56%), but also resulted in a significant change to the design, giving it a four-leaf clover-like shape rather than thin, petal-like spikes.

#09419

Another small step forward in efficiency (59 percent) and another striking design choice by MAMA BEAR. Created in January 2022, the shape resembles a butterfly with a narrowing downwards in the center.

#14125

If you thought MAMA BEAR couldn't get any more artistic, you'd be wrong. The experiment resulted in a starburst effect with an efficiency rating of 60 percent.

#16043

Now we are very close to breaking the record: as of August 2022, MAMA BEAR has produced a 2.5-centimeter-tall shape with 65 percent efficiency, almost a centimeter taller than the previous 16,042 designs.

#17636

We're getting closer: In October 2022, after 1,593 designs since our last featured design, this shape achieved 71 percent efficiency. And MAMA BEAR has taken it even higher, at 2.78 centimeters.

#21285

Here's a rendering of the record-breaking shape. Mama Bear built this on January 13, 2023, with 75 percent efficiency. It's 2.64 centimeters tall.

The record-breaking structure is quite different from what researchers expected: It has four points, is shaped like thin petals, and is taller and narrower than earlier designs.

“We have a ton of mechanical data here and we're excited to use it to learn lessons about design in general,” Brown says.

To find the most efficient shape, the robot designed and milled over 25,000 different structures, running multiple experiments.

Their vast data is already being used in the field for the first time to help design new helmet padding for U.S. Army soldiers. Brown, SNAP, and project collaborator Emily Whiting, an associate professor of computer science at Boston University's School of Humanities and Sciences, worked with the U.S. Army recently to conduct field tests to ensure helmets using their patent-pending padding were comfortable and adequately protective in impacts. The 3D structure used in the padding differs from the record-breaking section in that it is softer in the middle and has a lower height for increased comfort.

Wire frame

solid

Use the slider to toggle between a 3D digital rendering and the interior frame of Structure #07214, which is 56 percent efficient.

MAMA BEAR isn't Brown's only autonomous research robot. His lab has other “BEAR” robots that perform a variety of tasks. For example, Nano BEAR uses a technique called atomic force microscopy to study materials behavior at the molecular level. Brown is also working with ENG assistant professor of mechanical engineering Jörg Werner to develop another system, PANDA (short for Polymer Analysis and Discovery Array) BEAR, to test thousands of thin polymer materials to find the best ones for batteries.

“These are all robots doing research,” Brown says. “The philosophy is that by using a combination of machine learning and automation, we can greatly increase the speed of research.”

“It's not just fast,” Snap adds. “It can do things that you can't normally do. It can reach structures and goals that you couldn't reach any other way because it would be too expensive or too time consuming.” Snap has worked closely with Mama Bear since the experiment began in 2021, giving the robot machine vision and the ability to wash its own test plates.

KABlab wants to further demonstrate the importance of autonomous research. Brown wants to continue collaborating with scientists from different disciplines who need to test huge numbers of structures and solutions. Even though they've already broken a record, “we don't know if we've reached maximum efficiency,” Brown says. That means it could break the record again. So while MAMA BEAR will keep running and pushing its limits further, Brown and his team will consider what other uses the database could be useful for. They're also exploring how the more than 25,000 pieces can be unwound and reloaded into the 3D printer, so the material can be reused for further experiments.

“We will continue to study this system because, like many other material properties, mechanical efficiency can only be accurately measured through experimentation,” Brown says, “and using the self-driving lab will enable us to select the best experiments and run them as quickly as possible.”

The research was supported by the National Science Foundation and the U.S. Army.

View related topics:



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *